Re: [Grep-devel] [PATCH 0/4] search backend thread safety

[Paul Eggert]

Zev Weiss wrote:
> Given that *some* data (e.g. the dfa produced by GEAcompile() and altered in
> EGexecute()) is going to need to be replicated per-thread, it seemed simplest to
> me to just localize all (non-const) globals across the board

Although that may be simpler in the short run, surely the code should 
distinguish between objects that are read-only during execution (and can be 
shared among worker threads) from objects where each thread needs its own copy. 
This is not merely to help performance; it's to help programmers understand the 
code better.